KR102607446B1 - Heart failure diagnosis method by measuring ventricular ejection fraction - Google Patents

Abstract

An exemplary embodiment of the invention disclosed herein relates to a method of diagnosing heart failure by dividing the endocardial border in a cardiac image and measuring the ventricular contracture rate. The heart failure diagnosis method includes obtaining a magnetic resonance image of the heart of a subject, dividing the endocardial border of the heart in the magnetic resonance image using a first neural network, and dividing the endocardial border of the heart using a first neural network. It may include measuring the volume of the ventricle through the step of measuring the volume of the ventricle, measuring the ejection fraction of the ventricle through the measured volume of the ventricle, and determining whether the ejection fraction falls within a preset range. You can.

Description

Heart failure diagnosis method by measuring ventricular ejection fraction}

The present invention relates to a method of diagnosing heart failure by dividing the endocardial border in cardiac images and measuring the ventricular contracture rate. This invention includes results carried out with support from the Startup and Commercialization Support Project of the Ministry of SMEs and Startups. [10448495, Development of an automated model for heart disease diagnosis through comprehensive cardiac magnetic resonance imaging (MRI) analysis]

Among various molecular imaging techniques, magnetic resonance imaging (MRI) provides excellent anatomical imaging based on the interaction between molecules surrounding the tissue (lattice) and hydrogen atoms (protons). Because it can provide images, it is considered one of the most powerful and non-invasive diagnostic tools.

Recently, magnetic resonance imaging imaging technologies are being developed to quantitatively measure biophysical quantities in addition to existing anatomical tomography images. Representative biophysical quantities that can be measured with MRI include the self-recovery time of hydrogen atoms due to high-frequency pulses in human tissue, blood flow rate, diffusion, and perfusion, and can be measured non-invasively, safely, and accurately.

Among these, contrast between tissues can be created through the characteristics of each component of self-recovery, and this difference in recovery time is actively used as a biophysical quantity that can quantitatively evaluate tissue disease at the molecular level.

For example, through cardiac magnetic resonance imaging of the heart, the movements of the heart's systole and diastole can be seen and the overall heart function as well as local heart function can be evaluated.

However, additional work is required for diagnosis through these images.

In particular, heart failure can be caused by a variety of causes, with cardiovascular (coronary artery) disease (e.g., myocardial infarction, etc.) being the most common cause (about 2/3), and heart muscle (myocardial) disease (e.g., The main causes are unknown causes, cardiomyopathy due to genetic causes, myocarditis due to viral infection, etc.), high blood pressure, valves, etc. In addition, long-term rapid pulse (tachycardia) and persistent stress can cause reversible heart failure that improves when the cause is removed, and some anticancer drugs may also cause heart failure in proportion to the cumulative dose used.

The most common and important symptom of heart failure is shortness of breath (difficulty breathing). Shortness of breath is mainly caused by blood stagnation (congestion) in the heart, which increases the filling pressure of the ventricle, which causes blood to stagnate in the pulmonary blood vessels entering the heart, which can cause coughing. Additionally, because the heart cannot properly pump blood, fatigue and decreased exercise capacity occur.

Heart failure can be diagnosed through subjective symptoms of heart failure such as shortness of breath, fatigue, and decreased exercise capacity, as well as general tests such as blood tests, chest x-rays, and electrocardiograms.

In the case of the conventional invasive method (blood test), there is the inconvenience of having to visit the patient directly and perform a preliminary examination for diagnosis, and in the case of the conventional non-invasive method (chest x-ray), the anatomical location of the heart and the It is necessary to distinguish common areas for multiple images taken at various shooting angles, and therefore, there is a problem in that the diagnosis results are affected by the operator's skill level.
Prior patents include Japanese Patent Publication No. 2019-504659 (2019.02.21.), US Patent Application Publication US2020/0397313 (2020.12.24.), and Chinese Patent Publication 110363772 (2019.10.22.).

Therefore, the present invention aims to solve the above-mentioned problems.

One of the various tasks of the present invention is to propose a method for segmenting the endocardial border in the heart region through a deep learning-based learned neural network.

One of the various tasks of the present invention is to propose a method that can increase the accuracy of measuring ventricular volume through divided endocardial boundaries.

One of the various tasks of the present invention is to propose a method for measuring the ventricular contracture rate by measuring the volume of the ventricle based on the image at the end of the diastole and the image at the end of the systole during the cardiac cycle.

One of the various tasks of the present invention is to propose a method for diagnosing heart failure by comparing the measured ventricular contracture rate with a preset value.

One of the various tasks of the present invention is to propose a specific learning method of a neural network for segmentation of the endocardial border in the cardiac region to achieve a higher diagnosis rate.

One of the various tasks of the present invention is to propose a method of securing learning data for a neural network using MRI images used in cardiac diagnosis.

Various embodiments for solving the problem of the present invention include obtaining a magnetic resonance image of the heart of a subject and segmenting the endocardial border of the heart in the magnetic resonance image using a first neural network. and measuring the volume of the ventricle through the divided endocardial border and measuring the ejection fraction of the ventricle through the measured volume of the ventricle, and whether the ejection fraction falls within a preset range. Provides a heart failure diagnosis method through measuring ventricular contracture rate, which is characterized in that heart failure is diagnosed by determining.

The obtained magnetic resonance image may be an image at the end of diastole or an image at the end of systole during the cardiac cycle.

In the step of measuring the volume of the ventricle, the volume of the ventricle may be measured by dividing the ventricle into a plurality of pieces along the boundary of the divided endocardium and then adding up the volumes of the pieces.

In the step of measuring the contracture rate of the ventricle, the contracture rate of the ventricle may be measured based on the volume of the ventricle measured in the image at the end of diastole and the volume of the ventricle measured in the image at the end of systole.

In the step of measuring the contracting rate of the ventricle, the contracting rate of the ventricle can be measured through a model learned using a second neural network.

The first neural network and the second neural network may form different network structures.

The first neural network is a first network that calculates feature values through a convolution operation on pixel values in the input image, and a division result value of the input image size through a deconvolution operation on the calculated feature values. It may be configured as a second network.

The first and second networks have a mutually symmetrical structure according to the convolution or deconvolution operation step, and the second network synthesizes the output value of the convolution step of the first network corresponding to the deconvolution output value ( concatenate) to create a feature map.

An exemplary embodiment of the present invention includes an image acquisition unit that acquires a magnetic resonance image of the heart of a subject, and a region divider that divides the endocardial border of the heart from the acquired image using a first neural network. An ejection fraction measuring unit that measures the volume of the ventricle through the division and the divided endocardial border and measures the ejection fraction of the ventricle through the measured volume of the ventricle, and the measured ejection fraction is preset. A heart failure diagnosis device is provided by measuring ventricular contracture rate, comprising a diagnosis unit that determines whether the device falls within the range and diagnoses heart failure.

The acquired magnetic resonance image may be an image at the end of diastole and an image at the end of systole during the cardiac cycle.

The volume of the ventricle may be measured by dividing the ventricle into a plurality of pieces along the boundary of the divided endocardium and then adding up the volumes of the pieces.

The ventricular contracture rate may be measured based on the volume of the ventricle measured in the image at the end of diastole and the volume of the ventricle measured in the image at the end of systole.

The contracture rate of the ventricle may be measured through a model learned using a second neural network.

The first neural network and the second neural network may form different network structures.

An exemplary embodiment of the present invention includes obtaining a magnetic resonance image of the heart of a subject, segmenting the endocardial border of the heart in the magnetic resonance image using a first neural network, and dividing the endocardial border of the heart using a first neural network. Measuring the volume of the ventricle through the endocardial border, measuring the ejection fraction of the ventricle through the measured volume of the ventricle, and determining whether the ejection fraction falls within a preset range. Provided is a computer-readable recording medium storing a program for performing a method for diagnosing heart failure by measuring ventricular contracture rate, which includes the step of diagnosing heart failure.

Each feature of the above-described embodiments can be implemented in combination in other embodiments as long as it is not contradictory or exclusive to other embodiments.

According to various embodiments of the present invention, resources used for image interpretation can be reduced by accurately and quickly segmenting the boundaries of the endocardium in cardiac images using a neural network.

In addition, the present invention obtains unified segmentation results regardless of the skill level of the operator reading the image, thereby simplifying the diagnosis process and increasing the reliability of diagnosis.

In addition, the present invention diagnoses heart failure by estimating the ventricular contracture rate through accurately and quickly segmented endocardial boundaries using a neural network, thereby increasing the reliability of the diagnosis.

The effects of the present invention are not limited to those described above, and other effects not mentioned will be clearly recognized by those skilled in the art from the description below.

Figure 1 is a diagram schematically showing a system for diagnosing heart failure by measuring ventricular contracture rate according to an embodiment of the present invention.
Figure 2 is a diagram showing the heart cycle and electrocardiogram.
Figure 3 is a diagram showing a cross section of the heart.
Figure 4 is a diagram showing the flow of a method for dividing the border of the endocardium according to an embodiment of the present invention.
Figures 5 and 6 are diagrams illustrating the configuration of a neural network according to an embodiment of the present invention.
Figure 7 is a diagram showing the flow of a method for diagnosing heart failure by measuring ventricular contracture rate according to an embodiment of the present invention.
Figure 8 is a diagram showing the ventricle divided into a plurality of pieces along the border of the endocardium to measure the volume of the ventricle according to an embodiment of the present invention.
Figure 9 is a diagram showing the flow of a method for diagnosing heart failure by measuring ventricular contracture rate according to an embodiment of the present invention.

Hereinafter, specific embodiments of the present invention will be described with reference to the drawings. The detailed description below is provided to provide a comprehensive understanding of the methods, devices and/or systems described herein. However, this is only an example and the present invention is not limited thereto.

In describing the embodiments of the present invention, if it is determined that a detailed description of the known technology related to the present invention may unnecessarily obscure the gist of the present invention, the detailed description will be omitted. In addition, the terms described below are terms defined in consideration of functions in the present invention, and may vary depending on the intention or custom of the user or operator. Therefore, the definition should be made based on the contents throughout this specification. The terminology used in the detailed description is merely for describing embodiments of the present invention and should in no way be limiting. Unless explicitly stated otherwise, singular forms include plural meanings. In this description, expressions such as “comprising” or “comprising” are intended to indicate certain features, numbers, steps, operations, elements, parts or combinations thereof, and one or more than those described. It should not be construed to exclude the existence or possibility of any other characteristic, number, step, operation, element, or part or combination thereof.

Additionally, in describing the components of the embodiment of the present invention, terms such as first, second, A, B, (a), and (b) may be used. These terms are only used to distinguish the component from other components, and the nature, sequence, or order of the component is not limited by the term.

Magnetic resonance imaging (MRI), that is, MRI, uses an inversion recovery RF pulse to invert atoms magnetized by a magnetic field and then classifies the components of the recovery process that are recovered in the form of an exponential curve, and determines the recovery time according to the components. can be calculated. An MRI signal is acquired by scanning the signal generated from the subject in k-space, and the obtained MRI signal is converted to obtain an MRI image.

In the case of the heart, to evaluate the function of periodically beating ventricles, multiple images acquired through continuous MRI scans synchronized with the cardiac cycle can be reconstructed in each k-space and generated as cine images. there is.

In addition, MRI provides a variety of images by emphasizing signals by adjusting the variables (Tim to repetition, Time to echo, etc.) of RF (Radio Frequency) pulses that apply characteristics of each component of self-recovery according to the purpose of diagnosis. It is also possible to obtain it.

At this time, TR refers to the repetition time of the RF pulse. TR refers to the time interval for generating RF pulses used to obtain resonance signals, and mainly determines the amount of longitudinal relaxation (Spin-Lattice).

TE is the signal generation time and refers to the time from outputting the RF pulse to obtaining the echo signal. TE determines the degree of dispersion (dephase) of the spin that is recovered onto the transverse plane (Spin-Spin).

At this time, T1 is defined as T1 relaxation time, which is the time from when the RF pulse is injected and reversed until an average magnetization of 63% of the initial state is formed in the longitudinal direction.

T2 is defined as the time it takes for the average magnetization in the transverse plane to decrease to 37% of the initial value by dephasing.

In other words, a T1-weighted image or a T2-weighted image can be generated by measuring the above time by varying the TR and TE of the RF pulse.

Hereinafter, a method of performing artificial intelligence-based diagnosis through movie images will be described as an example of a method of diagnosing myocarditis by measuring ventricular contracture rate according to an exemplary embodiment of the present invention.

Figure 1 is a diagram schematically showing a system for diagnosing heart failure by measuring ventricular contracture rate according to an embodiment of the present invention.

Hereinafter, the description will be made with reference to Figure 1.

Referring to FIG. 1, the segmentation device 100 that performs an automatic segmentation method of the endocardial border area through a neural network in this embodiment may be configured in conjunction with the MRI device 1000.

As described above, the MRI device 1000 applies magnetic fields and RF pulses to the human body of the photographer 10 and extracts characteristics of the recovery process after magnetization and reversal in a specific axial direction of hydrogen atoms. The MRI device 1000 enables non-invasive image acquisition and diagnosis by generating MRI images using signals received during the recovery process.

The segmentation device 100 of this embodiment may be composed of an image acquisition unit 110 and an endocardial boundary area segmentation unit 130.

The image acquisition unit 110 may receive a magnetic resonance image captured by the MRI device 1000. The MRI device 1000 can calculate as a signal the component values of the spin-lattice recovery process in which atoms that have been longitudinally magnetized according to the direction of the magnetic field are inverted using an inversion recovery RF pulse and then recovered in the form of an exponential curve. Component values generated from the heart of the photographer 10 are scanned in k-space to obtain an MRI signal, the acquired MRI signal is mapped in pixel units, and a plurality of images acquired according to the cardiac cycle are reconstructed into an image to produce a movie image. creates .

The image acquisition unit 110 is also capable of receiving generated movie images or generating them directly.

In order to diagnose heart disease through acquired movie images, segmentation of areas such as the epicardium, endocardium, myocardium, and endocardial border in the heart within the image is important. Therefore, for complete quantitative analysis, a large number of images must be accurately segmented.

However, since this segmentation process is influenced by the reader's skill level and concentration, it is necessary to perform accurate segmentation based on a more common standard through automation.

Accordingly, in this embodiment, the endocardial border area segmentation unit 130 divides the inner border area of the endocardium using a movie image as an input to the neural network 200. The neural network 200 performs learning through various learning data for more accurate segmentation and thereby improves prediction performance in the above-mentioned areas.

The diagnostic device 300 can diagnose heart failure using information on the endocardial border area divided by the dividing device 100.

More specifically, the diagnostic device 300 may include a build rate measurement unit 310 and a diagnostic unit 330. The contracture rate measuring unit 310 can measure the ventricular ejection fraction based on the boundary of the endocardium divided through the region dividing unit 130.

The ventricular contracture rate is a value that can evaluate the systolic function of the ventricle, and is one of the indicators that can measure the degree to which blood flowing into the left ventricle flows out through the aorta during the ventricular contraction period.

In this embodiment, the ventricular ejection fraction may mean the left ventricular ejection fraction.

More specifically, heart failure is a condition in which the function of the heart is reduced due to various causes as described above, and as an example of the cause, myocarditis, a type of heart muscle disease, is a condition in which inflammatory cells infiltrate the heart muscle either acutely or chronically. , indicates damage to myocardial cells. In this case, heart function deteriorates and may develop into diseases such as dilated cardiomyopathy (DCM). Dilated cardiomyopathy is a type of heart contractile disorder in which blood accumulates within the heart and the atria or ventricles expand due to contractile dysfunction of the myocardium.

Therefore, in one embodiment of the present invention, the contracture rate of the left ventricle is measured as a scale for determining the function of the heart to diagnose heart failure and determine whether it falls within a preset range.

The ventricular contracture rate can be calculated by measuring the volume change of the ventricle through the boundary of the divided endocardium, which will be described in more detail with reference to the drawings below.

Heart failure can be diagnosed by determining whether the ventricular contracture rate measured through the contracture rate measurement unit 310 falls within a preset range in the diagnosis unit 330.

More specifically, the preset range can determine (binary diagnosis) whether the measured ventricular contracture rate falls within the normal range or abnormal range. Alternatively, it is possible to determine (category diagnosis) whether the measured ventricular contracture rate corresponds to a plurality of categories defined by a preset range.

In addition, the preset range can increase the diagnostic accuracy of heart failure by utilizing basic data of various modalities. As an example, it may refer to information such as gender, race, and image modality that affects basic data for diagnosing heart failure. Alternatively, it may reflect special information such as underlying disease, genetic testing, and family history of the photographer (10).

Figure 2 is a diagram showing the heart cycle and electrocardiogram.

Hereinafter, the description will be made with reference to FIG. 2.

Specifically, the bottom of FIG. 2 is a diagram showing an electrocardiogram (ECG) recorded by amplifying the electrical activity of the heart in response to the cardiac cycle. In general, one cycle of ECG is divided into PQRST waves, with the P wave due to atrial excitation, the QRS wave due to ventricular excitation, and the T wave due to the ventricular recovery process.

Figure 2(a) is a diagram showing the section where the P wave is observed, meaning the point in time when atrial contraction begins due to electrical stimulation. Figure 2(b) represents the section in which atrial contraction is in progress after the P wave is observed.

Figure 2 (c) is a diagram showing the section in which the QRS complex is observed, meaning the point in time when ventricular contraction begins due to electrical stimulation. The Q wave refers to the negative (downward) pole that occurs behind the P wave and in front of the R wave, and is the first downward pole of the QRS complex, which represents excitement of the mind and body. The R wave refers to an upward spike wave among the intraventricular excitation waves in an electrocardiogram. The S wave refers to the downward oscillation of the QRS complex that occurs following the R wave in a normal electrocardiogram. Figure 2 (d) is the section where the T wave is observed, which is formed by repolarization of the ventricular muscle.

That is, the heart goes through cycles such as initial diastole, atrial systole, atrial systole, ventricular ejection, and ventricular recovery, and the corresponding changes in heart volume and electrocardiogram are as shown in FIG. 2.

In this embodiment, an image of the end diastole and an image of the end systole during the cardiac cycle are obtained from a magnetic resonance image of the heart of the photographer 10, and the endocardial border of the heart can be divided using a neural network.

Left ventricular ejection fraction (LVEF) can be calculated using the formula below.

Formula: LVEF(%)=(diastolic volume of left ventricle - systolic volume of left ventricle)/diastolic volume of left ventricle*100

Since the systolic and diastolic images of the left ventricle change in time series, to accurately calculate LVEF, as described above, the left ventricular volume at the end of diastole, when the volume of the left ventricle is the largest, and the left ventricular volume at the end of systole, when the volume of the left ventricle is the smallest, are used as the standard. It is desirable to do so.

The end of the diastolic phase described above may refer to a temporary period in the process of transitioning from (b) in Figure 2 to (c) in Figure 2. In other words, the volume of the ventricle increases as the ventricle relaxes, and the period immediately before the ventricle contracts can be defined as the end of the tracheal period of the above-mentioned cardiac cycle.

And the end of systole described above may refer to a temporary period in the process of transitioning from (d) in Figure 2 to (a) in Figure 2. In other words, the volume of the ventricle decreases as the ventricle contracts, and the period immediately before the ventricle relaxes again can be defined as the end systole of the above-mentioned cardiac cycle.

Therefore, in this embodiment, in the step of segmenting the endocardial border of the heart using a neural network in the magnetic resonance image of the heart to measure LVEF, the magnetic resonance image is an image of the end diastole and an image of the end systole of the heart. can be used. By using images of the heart at the end of diastole and end systole, when measuring LVEF, it is possible to minimize changes in ventricular volume due to changes in ventricular volume during the cardiac cycle or other factors such as increase in myocardial thickness, which is the standard for measuring LVEF. Accurate measurement of left ventricular volume is possible.

In other words, the left ventricular volume at the end of diastole is measured by dividing the border of the endocardium in the image at the end of diastole during the cardiac cycle, and the left ventricle volume at the end of systole is measured by dividing the border of the endocardium in the image at the end of systole during the cardiac cycle. Through this, the LVEF can be measured and compared with a preset range to increase the accuracy of diagnosing heart failure.

Figure 3 is a diagram showing a cross section of the heart.

Hereinafter, the cardiac cycle, ECG, and heart signal transmission will be described in detail with reference to FIGS. 2 and 3. Blood that circulates throughout the body through the arterial system through the vena cava 91 is collected and enters the right atrium. After blood flows into the right atrium, an electrical signal is generated in the SA node (911), causing contraction of the right atrium, and the generated signal is transmitted through the AV node (912) and the Bundle of His (913) to Purkinje fibers ( Signals are transmitted through Purkinje fibers, 914).

More specifically, the heart 9 beats on its own, and there is a beating source in the heart 9 containing autonomous beating cells that periodically generate depolarization on their own.

Depolarization generated in autonomous beating cells is transmitted to contractile cells, and contractile cells transmit electrical signals to other contractile cells using intervening plates. The place that generates the initial beat is the sinoatrial node (SA node, 911).

Since the sinoatrial node 911 is located in the right atrium, and there is fibrous tissue with an insulating function between the atrium and the ventricle, a specialized transmission system must exist to transmit signals generated in the sinoatrial node 911.

Another node where autonomous beating cells gather to transmit the initial signal generated in the sinoatrial node to the ventricle is the atrioventricular node (AV node, 912). The electrical signal from the atrioventricular node 912 transmitted in this way is not directly transmitted to the contractile cells of the ventricle, but is transmitted to the apex of the heart through a different transmission system so that contractile cells at the apex of the heart (lower part of the heart) contract. must be conveyed.

The Bundle of His (913) is an intermediate electrical signal conduction system that transmits signals from the atrioventricular node (912) to the Purginae fibers (914) via the Bundle of His (913) below.

In this embodiment, among the magnetic resonance images of the heart of the subject described above, the boundary area of the endocardium is segmented from the images of the end diastole and end systole of the cardiac cycle, and the left ventricular volume at the end of diastole and end systole is measured. Heart failure can be diagnosed by measuring LVEF and determining whether it falls within a preset range.

Hereinafter, with reference to FIGS. 4 to 6, a method for segmenting feature regions in the above-described end-diastolic and end-systolic cardiac images will be described.

Figure 4 is a diagram showing the flow of a method for dividing the boundary of the endocardium according to an embodiment of the present invention, and Figures 5 and 6 are diagrams exemplarily showing the configuration of a neural network according to an embodiment of the present invention. .

First, referring to FIG. 4, the heart region segmentation method using the neural network 200 according to this embodiment will be described in more detail.

A movie image generated by mapping the recovery time according to the standard recovery rate of protons in heart tissue inverted with RF (Radio Frequency) pulses in pixel units in a two-dimensional space is input (S21).

As described above, a movie image is obtained as an MRI signal by scanning component values generated from the photographer's heart in k-space, and the obtained MRI signal is mapped in pixel units to generate a movie image.

At this time, when the recovery time for each tissue of the heart is measured in pixels, changes in the positions of the epicardium and endocardium due to the natural movement of the myocardium according to the heart rate may affect it, so in this embodiment, the recovery time for each tissue is measured. The signal can be acquired by synchronizing it with the heart rate cycle.

Next, using the acquired movie image, the segmentation result is output using a neural network 200 that segments the endocardial border area within the movie image (S23).

Referring to FIG. 6, in this embodiment, the neural network 200 may be implemented as a CNN (Convolution Neural Network) model composed of a stacked structure of layers that perform a convolution operation.

The convolution layers within the CNN model concatenate the feature values in the input movie image 210 and the filter within the layer, and through this, the features of the object to be segmented are emphasized and output. For the output feature map, the segmentation result 220 is output by outputting pixel-level object position information according to the input movie image size through an up-sampling process.

Specifically, in this embodiment, the neural network 200 is composed of a first network that performs convolution and a second network that performs up-sampling in order to output the segmentation result at higher resolution, but the steps of each network are mutually symmetrical. It can be configured.

The detailed structure of the network will be described in more detail below with reference to FIG. 7.

Referring to FIG. 7, the neural network 200 according to this embodiment includes a first network that calculates feature values through a convolution operation on pixel values in the input image, and a deconvolution operation on the calculated feature values. It may be configured as a second network that calculates a division result value of the input image size.

At this time, the first network gradually outputs a feature map in which the characteristics of the epicardial and endocardial regions are emphasized through a convolution operation for the input movie image 210. Additionally, the feature map can be reduced in size by Max-pooling and the number of channels can be expanded depending on the filter being calculated.

Finally, the feature map generated in the first network can be re-extended again through the second network having a mutually symmetrical structure.

The second network can expand the size of the image through de(up)convolution instead of reducing the number of channels.

At this time, each stage of the second network uses the image of an expanded size through deconvolution and the output value of the corresponding stage of the first network to perform a convolution operation and generate a feature map for each stage.

That is, the second network concatenates the deconvolution result and the corresponding output value of the convolution step of the first network, and performs a convolution operation on the synthesized image to generate a feature map.

The above process can also be repeated for each step corresponding to the first network, and preferably can be performed until the size of the input movie image and the size of the feature map output from the second network become the same. The final feature map of the second network can output a segmentation result of two regions. Specifically, in this embodiment, the neural network 200 divides the heart tissue into the inner and outer regions of the myocardium, and outputs the segmentation result as a segmentation map. It can be output as (220).

Furthermore, in this embodiment, the neural network 200 may configure intermediate images used for generating movie images other than the movie image of the end diastole as a learning data set in order to secure various learning data.

Hereinafter, learning of the neural network 200 according to this embodiment will be described.

In this embodiment, the neural network 200 learns using a movie image of end-diastole and end-systole used for diagnosis, or an image set consisting of an intermediate map obtained by taking multiple shots per heart beat cycle to generate the movie image. It can be.

Therefore, in this embodiment, a plurality of intermediate maps are obtained according to the heart beat cycle and RF pulse synchronization. As mentioned above, in the case of the heart, the exercise cycle of the heartbeat can be measured by measuring the electrical signals generated from the heart called an electrocardiogram (ECG). For example, the beat cycle can be determined based on a specific waveform among the ECG signals. there is. Specifically, by using the R wave, which represents the highest point in the QRS complex, to obtain an intermediate map with the RR interval, which is defined as the interval between the R wave and the R wave of the next signal, images can be acquired while minimizing the influence of heart rate.

Specifically, after an RF pulse is applied, signals at specific points in the RR interval can be measured multiple times. When the first inversion occurs after the first RF pulse is applied, a plurality of partial images can be acquired by repeatedly measuring the recovery signal in the longitudinal axis direction of atoms in the heart tissue according to the RR interval.

At this time, since the movie image and the intermediate maps used to generate the movie image are all images taken of the same heart, they are formed as one learning set and used as learning data for the neural network 200 that performs segmentation of the heart region.

In addition, it is of course possible to augment the learning data through additional image processing to diversify the characteristics of the image.

Figures 7 and 9 are diagrams showing the flow of a method for diagnosing heart failure by measuring ventricular contracture rate according to an embodiment of the present invention, and Figure 8 shows the flow of the endocardium for measuring the volume of the ventricle according to an embodiment of the present invention. This is a diagram showing the ventricle divided into a plurality of pieces along the border.

Hereinafter, description will be made with reference to FIGS. 7 to 9.

First, referring to FIG. 7, in this embodiment, a magnetic resonance image of the heart is acquired (S10) as described above. In this case, as described above, it is possible to more accurately measure the left ventricular contracture rate by acquiring the end-diastolic image and the end-systolic image of the heart during the cardiac cycle.

Based on the acquired image, the endocardial border of the ventricle can be segmented (S30) using a neural network. At this time, as described above, it is desirable to divide the divided area into the endocardial border area of the left ventricle for accurate diagnosis of heart failure.

After dividing the border of the endocardium using a neural network, the contracture rate of the left ventricle can be measured (S50), and diagnosis of cardiac dysfunction is made by determining whether the measured contracture rate falls within a preset range. By doing so, heart failure can be diagnosed (S70).

Measurement of the contracture rate of the left ventricle (S50) through the endocardial border area of the divided left ventricle will be described in more detail with reference to FIG. 8 below.

Figure 8 is a diagram showing the ventricle divided into a plurality of pieces along the border 933 of the endocardium to measure the volume of the ventricle according to an embodiment of the present invention. The border 933 of the endocardium may be divided between the endocardium and the ventricular chamber, and may be divided at a predetermined distance from the epicardium 931. A myocardium exists between the above-described border 933 of the endocardium and the epicardium 931.

The volume of the ventricle can be measured by dividing the ventricle into a plurality of pieces along the boundary 933 of the divided endocardium and then adding up the volumes of the pieces. The plurality of pieces form a predetermined area (a) along the width direction of the divided endocardial border 933, and form a predetermined height (h) along the height direction of the divided endocardial border 933. Therefore, the volume of the unit piece can be calculated as the product of the predetermined area (a) and the predetermined height (h). And the volume of the ventricle can be calculated by adding up the volumes of all calculated unit pieces.

Of course, as described above and the volume of the ventricle in this embodiment, the end-diastolic volume of the heart and the end-systolic volume of the heart can be calculated for more accurate measurement of the contraction rate of the left ventricle.

The left ventricular contracture rate can be measured as a percentage by dividing the calculated difference between the end-diastolic volume of the heart and the end-systolic volume of the heart by the end-diastolic volume of the heart. Heart failure can be measured by determining whether the measured left ventricular contracture rate falls within a preset range and measuring heart function disorders, which will be described in more detail with reference to FIG. 9 below.

Figure 9 is a diagram specifically showing the flow of a heart failure diagnosis method through measuring left ventricular contracture rate according to an embodiment of the present invention.

The heart failure diagnosis method of this embodiment first acquires image data about the heart (S110), extracts images of the end-diastole and end-systole of the cardiac cycle from the acquired images (S130), and then extracts the heart images of the extracted end-diastole and The heart image at the end of systole can be segmented (S310). In this case, the divided area may mean the endocardial border of the left ventricle of the heart in the image. And, based on the divided endocardial border, the contracture rate of the left ventricle can be measured (S50). As described above in FIG. 8, the left ventricle is divided into a plurality of pieces, and then the volumes of the pieces are added to obtain the ventricular volume. It can be measured by measuring.

The contracture rate of the left ventricle is a calculation of the amount of change and a prediction of the amount of change based on continuous images, so it can be measured through a model learned using LSTM (Long-short term memory) or RNN (Recurrent Neural Network, RNN)-based neural network.

RNN, or Recurrent Neural Network, is configured so that the state is remembered inside the neural network and the state information can be circulated and connected, making it possible to consider continuous input data. Therefore, RNN is an appropriate model when dealing with continuous data whose order is meaningful, such as cardiac image data and left ventricular contracture rate that change in time series according to the cardiac cycle. However, RNN inherently has a vanishing gradient problem, and a Long Short Term Memory (LSTM) network was proposed to solve this problem, so the learning model in this embodiment may be constructed through an LSTM-based neural network.

Meanwhile, it is possible to determine whether the measured contracture rate of the left ventricle falls within a preset range (S600) and to diagnose whether the left ventricle is dysfunctional (S700). Heart failure can be diagnosed depending on whether the left ventricle is dysfunctional.

It may be determined (S611) whether the measured contracture rate of the left ventricle falls within a preset first range. The first range may mean a range of 70% or more, and if the left ventricular contracture rate falls within the first range, hyperdynamics among cardiac dysfunctions can be confirmed (S711).

If the measured contracture rate of the left ventricle does not fall within the first range, it may be determined whether it falls within the second range (S612). The second range may mean a range of 50% to 70%, and if the left ventricular contracture rate falls within the second range, normal without cardiac dysfunction can be confirmed (S712).

If the measured contracture rate of the left ventricle does not fall within the second range, it may be determined whether it falls within the third range (S613). The third range may mean a range of 40% to 49%, and if the left ventricular contracture rate falls within the third range, mild dysfunction among cardiac dysfunctions can be confirmed (S713). You can.

If the measured contracture rate of the left ventricle does not fall within the third range, it can be determined whether it falls within the fourth range (S614). The fourth range may mean a range of 30% to 39%, and if the left ventricular contracture rate falls within the fourth range, moderate dysfunction among cardiac dysfunctions can be confirmed (S714). You can.

If the measured contracture rate of the left ventricle does not fall within the fourth range, it can be determined whether it falls within the fifth range (S615). The fifth range may mean a range of less than 30%, and if the left ventricular contracture rate falls within the fifth range, severe dysfunction among cardiac dysfunctions can be confirmed (S715). The above-mentioned hyperdynamics, early dysfunction, severe dysfunction, and severe dysfunction can be used as one of the indicators for diagnosing myocarditis.

As described above, in order to determine heart failure, it is possible to determine (binary diagnosis) whether the ventricular contracture rate falls within the normal range (the second range), or it can be determined as a preset range as described above. It is also possible to determine (category diagnosis) whether or not it falls into multiple categories.

Although various embodiments of the present invention have been described in detail above, those skilled in the art will understand that various modifications can be made to the above-described embodiments without departing from the scope of the present invention. . Therefore, the scope of the present invention should not be limited to the described embodiments, but should be determined by the patent claims described later as well as equivalents to these claims.

10: Photographer 1000: MRI device
100: Splitting device 200: Neural network
300: Diagnostic device 400: Data storage unit

Claims (15)

In a method for outputting heart failure determination results through measurement of ventricular contracture rate performed by a computing device,
Acquiring magnetic resonance images of the subject's heart at the end of diastole and end systole during the cardiac cycle;
Segmenting the endocardial border of the heart in the magnetic resonance image using a first neural network;
measuring the volume of the ventricle through the divided endocardial border and measuring the ejection fraction of the ventricle through the measured volume of the ventricle; and
A step of determining the degree of cardiac dysfunction by determining whether the contracture rate falls within a preset range,
The first neural network is trained using an image set consisting of an intermediate map obtained as a unit of the R waveform of the electrocardiogram (ECG) signal and the interval (RR interval) between the R waveforms measured in the next signal,
The image set is a plurality of partial images obtained by measuring a plurality of signals at the time when the first inversion occurs after an RF (Radio Frequency) pulse is applied within the RR interval,
The step of measuring the contracting rate of the ventricle is calculated using a second neural network that predicts the contracting rate of the ventricle based on time-series volume changes of unit pieces that divide the area divided by the first neural network into a plurality of pieces. A method for outputting heart failure judgment results by measuring ventricular contracture rate. delete According to paragraph 1,
In the step of measuring the volume of the ventricle, the volume of the ventricle is measured by dividing the ventricle into a plurality of pieces along the boundary of the divided endocardium and then adding up the volumes of the pieces. Method for outputting heart failure judgment results through rate measurement. According to paragraph 1,
In the step of measuring the contracture rate of the ventricle, the contracture rate of the ventricle is measured based on the volume of the ventricle measured in the image at the end of diastole and the volume of the ventricle measured in the image at the end of systole. Method for outputting heart failure judgment results by measuring ventricular contracture rate. delete According to paragraph 1,
A method for outputting heart failure determination results through measuring ventricular contracture rate, characterized in that the first neural network and the second neural network form different network structures. According to paragraph 1,
The first neural network is a first network that calculates feature values through a convolution operation on pixel values in the input image, and a division result value of the size of the input image through a deconvolution operation on the calculated feature values. A method for outputting heart failure determination results through measurement of ventricular contracture rate, characterized in that it consists of a second network that calculates. In clause 7,
The first and second networks have a mutually symmetrical structure depending on the convolution or deconvolution operation step,
The second network generates a feature map by concatenating the deconvolution output value and the corresponding output value of the convolution step of the first network. An image acquisition unit that acquires magnetic resonance images of the subject's heart at the end of diastole and end systole during the cardiac cycle;
a region dividing unit that divides the endocardial border of the heart from the acquired image using a first neural network;
an ejection fraction measuring unit that measures the volume of the ventricle through the divided endocardial border and measures the ejection fraction of the ventricle through the measured volume of the ventricle; and
It includes a diagnostic unit that determines the degree of cardiac dysfunction by determining whether the measured contracture rate falls within a preset range,
The first neural network is trained using an image set consisting of an intermediate map obtained as a unit of the R waveform of the electrocardiogram (ECG) signal and the interval (RR interval) between the R waveforms measured in the next signal,
The image set is a plurality of partial images obtained by measuring a plurality of signals at the time when the first inversion occurs after an RF (Radio Frequency) pulse is applied within the RR interval,
The contracting rate measurement unit determines the contracting rate of the ventricle using a second neural network that predicts the contracting rate of the ventricle based on time-series volume changes of unit pieces that divide the area divided by the first neural network into a plurality of pieces. A device for outputting heart failure determination results through measurement of ventricular contracture rate, characterized in that: delete According to clause 9,
The volume of the ventricle is measured by dividing the ventricle into a plurality of pieces along the boundary of the divided endocardium and then summing the volumes of the pieces. According to clause 9,
The ventricular contracture rate is measured based on the volume of the ventricle measured in the image at the end of diastole and the volume of the ventricle measured in the image at the end of systole. Heart failure determination result output through measurement of ventricular contracture rate. Device. delete According to clause 9,
A device for outputting heart failure determination results through measurement of ventricular contracture rate, characterized in that the first neural network and the second neural network form different network structures. Acquiring magnetic resonance images of the subject's heart at the end of diastole and end systole during the cardiac cycle;
Segmenting the endocardial border of the heart in the magnetic resonance image using a first neural network;
measuring the volume of the ventricle through the divided endocardial border;
Measuring the ejection fraction of the ventricle through the measured volume of the ventricle; and
A step of determining the degree of cardiac dysfunction by determining whether the contracture rate falls within a preset range,
The first neural network is trained using an image set consisting of an intermediate map obtained as a unit of the R waveform of the electrocardiogram (ECG) signal and the interval (RR interval) between the R waveforms measured in the next signal,
The image set is a plurality of partial images obtained by measuring a plurality of signals at the time when the first inversion occurs after an RF (Radio Frequency) pulse is applied within the RR interval,
The step of measuring the contracting rate of the ventricle is calculated using a second neural network that predicts the contracting rate of the ventricle based on time-series volume changes of unit pieces that divide the area divided by the first neural network into a plurality of pieces. A computer-readable recording medium storing a program for performing a method for outputting heart failure determination results through measuring ventricular contracture rate, characterized in that:

Images